Measurements in the wireless

Transcription

1 Contents Measurements in the wireless Lecture slides for S Mika Ilvesmäki Motivation Wireless environment spesifics Wireless trace collection Wireless measurement methods Mobility measurements Networking laboratory Goals of this lecture After this lecture you should Know how to characterize the wireless environment in regards to measurements Know why measurements in the wireless are hard and how the gathering of measurement information is distributed Know where the wireline and wireless measurements and analysis are the same and where they differ Why measure? Wireless networks differ from the usual Internet access The access is (slightly) spatially dynamic Ad-hoc networking is easier Lack of wiring (=wireless) can create very highly populated networks Medium (the air interface) access is more complex and dynamic Clients and applications may show much more diversity 1

2 The same ol same ol Wireless networks contain the same users with same applications as networks in the wireline However, there might also be (a significant amount of) legacy customers from mobile voice networks (GSM). Same measurements and and analysis apply on layers 3 and above Same results definitely do not! Due to access layer being slightly different also the results may turn out to be slightly different Network traffic properties as a function of layer 2?! What is different? A network is a network However, WLAN performance is influenced 1. Mobility Handoff, handover 2. Channel noise 2,4/5GHz area is unlicensed, noise may be a problem with increased distance 3. Multiple access contention Access quality decreases as user count increases Measurements in layers 1 and 2 Transmission rates, retransmissions, signal strength Collision avoidance, detection RTS/CTS Traffic in different channels Wireless traces Incomplete Different depending on the measurement point Design is distributed, control points do not exist Inaccurate due to more probable configuration errors Subject to layer 2 features Contention very much dependent on where the measurement point is located and how it is moving (mobility) Channel usage and contention is location-dependent 2

3 Trace collection methods - nodes Wireless monitors Monitor all activity in the channels Event-triggered logs Memory and storage requirements Form a view (analysis!) of the network status and notify Other nodes in the WLAN Access point(s) Information/state distribution uses up bandwidth Careful planning and management of the measurement arrangements Do not overload production networks Do not lose relevan information Trace collection methods - BS Network trace dumps at the base station Poll access points with SNMPprotocol Gather info on amounts of connected users, transmission errors, and transmitted traffic Trace collection methods - Core Network trace dumps at aggregating routers Where traffic from several access points is aggregated Capture also traffic directed to WLANs Addtional monitoring and capturing load Determine traffic mix, network service profile etc. Normal traffic analysis Packet loss Two sources for packet loss Competing transimissions Miscalculation in power management Insufficient signal strength Probing for bandwidth Available BW depends on Amount of users in the coverage area Overall traffic load Available BW varies a lot Actively probing for BW introduces additional traffic and interferes with other traffic 3

4 Unintrusive bandwidth probing Configure dedicated queues for probe packets Forward probe packets only if no other traffic exists Larger packets/lower bandwidth still cause some interference to other traffic Egress router measures the packet arrivals and sends the info back to the sender Losses in wireless High error/loss rates Competing transmissions Signal strength related losses Adaptive transmission rates result in busier medium A lot of time data is sent at 1Mbps -> medium is reserved and contented and more losses/errors occur Multi-path fading produces a lot of (bit) errors Packets/bits received multiple times Packet delay WLANs produce larger transport delays Signal propagation (not that large contributor) Aggregating signal propagation results is significant delay Handovers Retransmissions Inference Jitter in wireless Due to medium dynamics jitter is much more pronounced in wireless Handovers Inferences, shadowing Link Layer Scheduling RTS/CTS 4

5 Measuring for mobility How often and how much a user moves while on line Physical movements are hard to measure RFIDs, but way-y in the future Detect roaming patterns from APs With dense AP placement, the node may change APs even if it is not moving Group APs into location clusters (per building, for instance) Mobility measures Time spent in one AP (location cluster) Movement speed (Mobile) session duration And visited APs in that time Application profile relation to movement Handoff measurements Handoff is performed when a mobile node transfers from one AP to another. Hand-off time is divided into two components Inter-cell time (rendezvouz-time), the time elapsed moving from one APs range to another In a well planned network this time should be negligible Hand-off setup time The time required to restore traffic flow when APs have changed Routing measurements in wireless Wireless environments are dynamic Especially ad-hoc networks Dynamic/Adaptive routing is mandatory Routing table sizes Rate of changes in the routing table Modeling Convergence times / dynamism of the network Modeling 5

6 Results from previous measurements High overhead Results show that only 40% of traffic is original data packets Retransmissions, acknowledgements, management traffic Retransmission rates are high (28% of data, 46% of data transmission time) Design for location access, not complete coverage Make sure you have APs where the users are, but not necessarily on the way to get there Users seem not to be moving when using WLANs Open&interesting issues User mobility modeling How do users move and how to model it? Inference studies with varying network topology More studies in wireless environments needed Application behavior in WLAN, 3G environments Overall traffic profiling and analysis This lecturer has no knowledge of analysis performed, for instance, in Aalto-network at TKK. Sell the idea, secure funding and get a summer/m.sc. Thesis job!! Ad Hoc networks will produce interesting analysis targets Summary of wireless/mobility measurements Several methods of gathering the traces No single method provides complete overview of the network status User profiling and characterization is twice as hard In addition to application usage there is also user mobility to account for Networks are also dynamic Mobile nodes, dynamic routing Distributed network status In nodes, APs and aggregate routers Several factors in various layers effect the network status Wireless network analysis is challenging No easy conclusions, no easy analysis 6